Vehicle Vectors and Traffic Patterns from Planet Imagery

TL;DR

The paper tackles the problem of deriving moving vehicle vectors from Planet imagery by leveraging the rainbow effect produced when sequential spectral bands are captured. It introduces two labeled datasets and a concurrent data framework to support large-scale, automated vehicle localization and velocity estimation across dual Planet constellations, SkySat and PlanetScope. Using a segmentation-plus-inference pipeline, it extracts speed and heading for moving vehicles and constructs vector fields, validated by SkySat and PlanetScope results and demonstrated through time-series analysis in Khartoum during a period of civil unrest. The work enables near-daily, global-scale traffic-pattern analytics and points to future improvements with raw L0 imagery for finer speed estimation and broader outlier analysis.

Abstract

We explore methods to detect automobiles in Planet imagery and build a large scale vector field for moving objects. Planet operates two distinct constellations: high-resolution SkySat satellites as well as medium-resolution SuperDove satellites. We show that both static and moving cars can be identified reliably in high-resolution SkySat imagery. We are able to estimate the speed and heading of moving vehicles by leveraging the inter-band displacement (or "rainbow" effect) of moving objects. Identifying cars and trucks in medium-resolution SuperDove imagery is far more difficult, though a similar rainbow effect is observed in these satellites and enables moving vehicles to be detected and vectorized. The frequent revisit of Planet satellites enables the categorization of automobile and truck activity patterns over broad areas of interest and lengthy timeframes.

Figures (15)

• Figure 1: Concurrent, co-located SkySat (top) and PlanetScope (bottom) imagery of a parking lot in Long Beach, CA.
• Figure 2: Concurrent SkySat (top) and PlanetScope (bottom) imagery of moving vehicles. Note the rainbow effect of moving vehicles in both images.
• Figure 3: SkySat car labeling schema. Static cars are labeled with a magenta dot, with moving cars labeled with green linestring
• Figure 4: Concurrent SkySat (left) and PlanetScope (right) imagery of moving vehicles taken over Perth, Australia with $\Delta t < 1 \, {\rm s}$. Top Left: Raw SkySat image. Top Right: Raw PlanetScope image. Bottom Left: SkySat image with labeled vehicles overlaid. "0" is a large vehicle (bus or truck), whereas "1", "2", "3" correspond to moving cars. "1" actually constitutes two unique cars, though we combine these into one label since these will be indistinguishable in PlanetScope; such labels are not used for SkySat training. Bottom Right: PlanetScope image with labeled vehicles overlaid.
• Figure 5: SkySat images and training masks for car detection. Left: Training image. Middle: Multi-class training mask with a 0.75 meter buffer around labels; magenta dots are static cars, and cyan lines are moving vehicles. Right: Training mask overlaid on the training image.